AI moderation

Moderation that runs by itself. Reviewable, reversible, by your rules.

The AI looks at each pair the way a reviewer would, says what speaks for and against a copy, and marks the pair when it is sure enough. Your people review its work and give the verdict.

Three levels

Hand over as much as you are comfortable with.

  1. Level 1

    A second opinion

    A moderator asks the AI about a pair or a whole product. It answers with a score and reasons, and the moderator marks.

  2. Level 2

    Automatic checks

    Every new match is checked without anyone asking. Moderators open a queue where the AI's view is already there, and apply it to many pairs at once.

  3. Level 3

    Autonomous moderation

    The AI marks the pairs it is sure about by itself: offender, suspect or cleared. Your team reviews its marks and handles only what it left open.

Safeguards

Automation you can check and take back.

You choose the designs

Autonomous moderation is switched on per protected product. Sensitive designs can stay with people.

You set how sure it must be

A threshold of its own for offender, suspect and cleared. Below it, the pair is left to a person.

Every AI mark is labelled

The history shows that the AI marked, when, and why. A filter lists the marks nobody has reviewed yet.

Confirm or take back

Reviewers confirm AI marks or change them. Marks no person has touched can be taken back all at once.

It has to earn “cleared”

The AI may clear pairs by itself only after it has proved on blind pairs that it rarely misses an offender.

It pauses by itself

If your team overrides the AI too often, or a daily limit is reached, it stops marking and tells you.

Measured against your team

ArxLegis counts missed offenders and false alarms, on pairs your people marked without seeing the AI's view.

A budget you know

The number of AI checks per month is agreed. The app shows how many are left, and every check is logged.

The verdict stays human

The AI marks. Whether a product is an offender in the end is decided by the people you gave that right.

Questions about the AI

Does the AI decide anything?

It marks, if you switch that on. The verdict always comes from a person, and every mark can be changed.

What does the AI see?

The pictures and texts of the two products of a pair, and your text about what is protected.

Where is it processed?

A profile with data residency in the EU is available.

Can we work without it?

Yes. AI moderation is a module of its own. Detection and the whole workflow work the same without it.

See it on your own catalog.

We set ArxLegis up with a few designs you have to respect and show you what looks similar in your assortment.

Book a demo